An AI agent sent about 300 emails during one office move

By Priya Raman

A San Francisco-based startup founder estimates that OpenAI's Dot agent probably sent around 300 emails in a few days while coordinating his office move, according to Business Insider.

Robert Farlow, 27, co-founded the AI startup Plato. He had just raised a new fundraising round and was growing the team. The company needed a bigger office. Building managers required insurance when movers arrived, while the new lease required coverage for compliance. The insurance and moving arrangements made the expansion operationally messy.

Farlow does not know much about insurance, so he would have consulted Claude or ChatGPT anyway. He uses AI agents throughout his company. Grok Bot takes the company's snack orders and buys them on Amazon.

He found an insurance contact and gave Dot a simple instruction. Farlow said: "We're moving to this office. You can read my emails to find out the details. Just get it done." Dot also texted him a couple of times. Farlow repeated the demand to get it done.

What did Dot do to the moving coordination?

Dot worked for a few days while the movers remained unconfirmed. It sent back-to-back messages, including triple and quadruple emails, and relayed every small update. When movers asked about a 9 a.m. start, Dot followed up to ask whether the time was Pacific Time. It needed clarification on nearly everything.

At times, Dot emailed the movers, the insurance contacts, and both the current and new building managers together, while another thread became too large for the new manager to open.

Dot also disclosed information about other parties. It told the movers about Brian, the building manager, although they had no need for that detail. The recipients initially did not know they were dealing with AI. The movers caught on and said they lacked the bandwidth to answer the agent's questions. Farlow believes the other parties never did, possibly assuming that an annoying person was spamming them.

Farlow estimates the agent probably sent around 300 emails in total over those few days. His as-told-to account, edited for length and clarity, gives no independent message count. OpenAI did not respond to Business Insider's request for comment.

What changed after the email flood?

When Farlow saw the movers' response, he asked: "What did I do?" He shut down Dot immediately. He has not replied to the movers yet, but wants to because he believes the agent put them through a lot.

The relationship with those movers ended regardless of the agent, because the company did not carry the insurance required for the job at that time. Farlow found another moving company and did not use Dot this time. He still uses AI to draft emails. Now, he reviews each draft before sending.

What does this change for AI users?

For an individual using AI to coordinate vendors, the practical change is to treat every external message as a draft requiring careful human review before sending. Farlow's experience shows the cost of delegating without a person checking the output: hundreds of emails, unnecessary disclosure and a strained vendor relationship.

The account offers no hiring, salary, layoff or productivity data. It cannot establish a national workplace trend. Its evidence concerns one founder, one office move and one agent, which is a useful operating lesson but not a measure of labor-market change.

Farlow expects the models to improve and more people to encounter similar behavior. He also expects office administration to shift toward an agent responding to another agent. His condition is practical: unless a user is sure the model will perform well, that person should review everything the agent produces.

Looking ahead, Farlow expects agents to answer other agents in office administration, while human review remains his safeguard as the models improve.